See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/286166879

Comparison of Forecasting Volatility in the Czech Republic Stock Market

Article  in  Applied Economics and Finance · December 2014

DOI: 10.11114/aef.v2i1.608

CITATIONS READS

11 569

3 authors:

Eleftherios Ioannis Thalassinos Erginbay Ugurlu

University of Piraeus Istanbul Aydin University
120 PUBLICATIONS   668 CITATIONS    47 PUBLICATIONS   108 CITATIONS   

SEE PROFILE SEE PROFILE

Yusuf Muratoglu
Hitit University
14 PUBLICATIONS   60 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Eastern Europe Economies View project

Muslim Economies View project

All content following this page was uploaded by Erginbay Ugurlu on 07 January 2016.

The user has requested enhancement of the downloaded file.

 Applied Economics and Finance
Vol. 2, No. 1; February 2015
ISSN 2332-7294 E-ISSN 2332-7308
Published by Redfame Publishing
URL: http://aef.redfame.com

Comparison of Forecasting Volatility in the Czech Republic Stock Market

Eleftherios I. Thalassinos1, Erginbay Ugurlu2 & Yusuf Muratoglu3
1
Professor, University of Piraeus, Department of Maritime Studies, Greece
2
Assistant Professor, Hitit University FEAS, Department of Economics, Turkey
3
Research Assistant, Gazi University FEAS, Department of Economics, Turkey
Correspondence: Eleftherios I. Thalassinos, University of Piraeus, Department of Maritime Studies, Greece.

Received: September 22, 2014 Accepted: October 13, 2014 Available online: December 11, 2014
doi:10.11114/aef.v2i1.608 URL: http://dx.doi.org/10.11114/aef.v2i1.608

Abstract
The aim of this paper is to examine different GARCH models with three different distributions in order to compare their
forecasting power in terms of volatility existing in the returns of the Czech Stock Market and more specific in the PX
index, for the period 08.01.2001-20.07.2012. We have employed GARCH, GJR-GARCH and EGARCH models against
normal, student-t and generalized error distributions. Then, we have forecasted stock market volatility for the Czech
Republic by its returns using the same models, GARCH, GJR-GARCH and EGARCH comparing their forecasting
performance. The results show that return volatility can be characterized by significant persistence and asymmetric
effects. We have estimated the corresponding variances for all models for the full sample period using static forecasts.
After comparing the forecasting performance of all nine models it was found that the EGARCH model has the best
forecasting performance compared to others.
Keywords: GARCH models, stock market volatility, forecasting performance
JEL classification: G15, G17
1. Introduction
Most financial data have some features which are leptokurtosis, volatility clustering or pooling and leverage effects.
Linear structural models and time series models are unable to explain some of the above important features.
Leptokurtosis, volatility clustering or pooling and leverage effects are tendencies for financial asset returns. The
tendencies of these features are defined as: a) to have distributions that exhibit fat tails and excess peakedness at the
mean; b) to have volatility in financial markets to appear in bunches which means large returns, of either sign and/or
small returns, of either sign; c) to have volatility to rise more following a large price fall than following a price rise of
the same magnitude (Brooks, 2008).
The most popular non-linear financial models are the Autoregressive Conditional Heteroscedastic (ARCH) or
Generalized Autoregressive Conditional Heteroscedastic (GARCH) which have been used for modeling and forecasting
volatility proposed by Engle (1982) and Bollerslev (1986).
The aim of this paper is to forecast stock market volatility for Czech Republic by its returns using GARCH,
GJR-GARCH and EGARCH models and compare their forecasting performance. The rest of the paper is organized as
follows. The second section is a literature review of related studies using GARCH models and stock market volatility.
The third section gives brief information about ARCH/GARCH models while the estimation results are presented in the
fourth section. The fifth section summarizes and concludes the paper.
2. Literature Review
Andersen, Bollerslev and Lange (1999) stated that the expected future volatility of financial market returns is the main
ingredient in assessing asset or portfolio risk and plays a key role in derivative pricing models. Emerson, Hall and
Zelweska-Mitura (1997) have investigated the Bulgarian stock market while Scheicher (1999) have examined Polish
stock returns. Shields (1997) modeled the returns of the Warsaw and Budapest stock exchanges returns. Scheicher
(2001) have characterized Hungary, Poland, and Czech markets as principal emerging stock markets in Europe. The
author has estimated a VEC model for each market modeling its volatility with a multivariate GARCH (M-GARCH)
structure. The findings show that the investigated countries have limited interaction and their volatility has a regional

11
Applied Economics and Finance Vol. 2, No. 1; 2015

character.
Vošvrda and Žıkeš (2004) have studied the behavior of volatility and the distributional properties of the Czech,
Hungarian, and Polish stock markets using weekly data for the period 1996-2002. They have used the PX-50 index for
Czech Republic and have found statistically significant results for GARCH (1,1) model concluding that the volatility of
the returns on PX-50 is very persistent.
Syriopoulos (2007) has investigated the relationships between the emerging stock markets, which are Czech Republic,
Hungary, Poland, Slovakia and two developed stock markets, Germany and the USA over the period 1997-2003.
Haroutounian and Price (2001) have investigated Czech Republic, Hungary, Poland, and Slovakia using both univariate
and multivariate GARCH models, namely GARCH, NGARCH, EGARCH, GJR-GARCH, AGARCH, NAGARCH and
VGARCH. The research has two parts which are the univariate models and the multivariate models. Relying on the
results of the univariate models, they have concluded that strong GARCH effects are apparent for all four markets with
the exception of the Czech Republic, where the coefficient of the lagged squared returns is not significant in three out of
seven specifications of conditional volatility.
Hajek (2007) has examined the Efficient Market Hypothesis on the Czech capital market for the period 1995–2005
using monthly, weekly and daily data. In this research, the author has analyzed efficiency and linear dependency of
several indices closing values and stock closing prices on the Prague Stock Exchange. He has concluded that both daily
stock returns and daily index returns are significantly linearly dependent, so the heteroskedasticity-consistent
methodology must be therefore applied to avoid significant biases.
Rockinger and Urga (2000) have investigated two groups of countries, which are the transition economies and the
established economies. Transition economies are Czech Republic, Poland, Hungary and Russia while the established
economies are USA, Germany and UK. Although they have focused on a sample of Central and Eastern European
Financial Markets (CEEFM) 1 they have only used these four countries. The model results are very similar for the
Czech Republic, Hungary and Poland while for all countries investigated significant GARCH effects have been
appeared.
Thalassinos and Politis (2011) have analyzed international stock markets for a number of countries using cointegration
analysis. They have found interrelationships between the selected variables a result that is consistent with other similar
studies.
Thalassinos, Maditinos and Paschalidis (2012) have observed significant evidence regarding insider trading in the
Athens Stock Exchange in the period 2000-2011.
3. Methodology
We have used four different types of GARCH models, ARCH, GARCH, GJR-GARCH and EGARCH structures as
follows:
3.1 ARCH Model
ARCH model is based on the variance of the error term at time t and depends on the realized values of the squared error
terms in lagged periods. The model is specified as in equations (1) to (3):

yt  u t (1)

u t ~ N0, h t  (2)
q
h t   0    j u 2t i (3)
t 1

This model is referred as ARCH(q), where q refers to the order of the lagged squared returns included in the model. If
we use ARCH(1) model equation (3) becomes:
h t   0  1u 2t 1 (4)
Since h t is a conditional variance, its value must always be strictly positive; a negative variance at any point in time
would be meaningless. To have positive conditional variance estimates, all of the coefficients in the conditional variance
are usually required to be non-negative. Thus coefficients must satisfy the condition 1  0 .
3.2 GARCH Model
Bollerslev (1987) and Taylor (1986) developed the GARCH(p, q) model. The model allows the conditional variance of

1 Czech Republic, Poland, Hungary, Russia, Bulgaria, Slovenia, Romania, Croatia and Estonia

12
Applied Economics and Finance Vol. 2, No. 1; 2015

the variable to be dependent upon previous lags; first lag of the squared residual from the mean equation presents news
about the volatility from the previous period which is as follows:
q p
h t   0    i u 2t i    i h t i
(5)
i 1 i 1

In the literature the most used and the simplest model is the GARCH(1, 1) process, for which the conditional variance
can be written as in equation (6):

h t   0  1u 2t 1  1h t 1 (6)

Under the hypothesis of covariance stationarity, the unconditional variance h t can be found by taking the
unconditional expectation of equation (6). Thus:
h   0  1h  1h (7)
Solving equation (6) we have:
0 (8)
h
1  1  1
For the existence of this unconditional variance it is required to have 1  1  1 and in order to be positive it must be
0  0 .
3.3 GJR GARCH Model
The GJR model is a simple extension of GARCH with an additional term added to account for possible asymmetries
(Brooks, 2008). Glosten, Jagannathan and Runkle (1993) developed the GARCH model which allows the conditional
variance to have a different response to past negative and positive innovations as shown in equation (9):
q p
h t   0    i u 2t i   i u 2t i d t 1    j h t  j (9)
i 1 i 1

Where is a dummy variable with values for bad and good news:

 1 if u t 1  0, bad news
d t 1  
0 if u t 1  0, good news

In this model, the effect of good news shows their impact by  , while bad news shows their impact by    . In
addition if   0 news impact is asymmetric and   0 leverage effect exists. To satisfy non-negativity condition
coefficients would be  0  0 ,  i  0 ,   0 and  i   i  0 . That is the model is still acceptable, even if
 i  0 , provided that  i   i  0 (Brooks, 2008).
3.4 Exponential GARCH MODEL
Exponential GARCH (EGARCH) proposed by Nelson (1991), has formed leverage effects in the equation. In EGARCH
model the specification for the conditional covariance is given by the following equation (10):

 
q p r
u t i ut k
log ht    0    j log ht  j    i   k (10)
j 1 i 1 ht i k 1 ht k

The two advantages stated in Brooks (2008) for the pure GARCH specification are: a) by using log ht even if the  
parameters are negative the equation will be positive; b) asymmetries are allowed under the EGARCH formulation.
In equation (10),  k represents leverage effect, which accounts for the asymmetry of the model. While the basic
GARCH model requires these restrictions, the EGARCH model allows unrestricted estimation of the variance (Thomas
and Mitchell, 2005).
If  k  0 , this is an indication that leverage effect exist, where  k  0 indicates an asymmetric impact in the model.
The meaning of the existence of leverage effect because of bad news increases the volatility.
When applying GARCH models to return series, it is often found that GARCH residuals still tend to be heavy tailed. To
accommodate this, rather than to use normal distribution the Student’s t and GED distribution have to be used to employ
ARCH/GARCH type models (Mittnik, Paolella and Rachev, 2002, Nelson, 1991).

13
Applied Economics and Finance Vol. 2, No. 1; 2015

3.5 Empirical Application

We have used daily data in stock exchanges of Czech Republic PX2 for the period 08.01.2001-20.07.2012. Statistical
data have been collected from Reuters. We have used the return term which is defined in Yu (2002) as follows:

 x 
r  log t  (11)
 x t 1 

Where is the capital index.

Graph 1 shows PX and its returns which is defined as RPX.

2,000 .15

.10
1,600
.05

1,200 .00

-.05
800
-.10
400
-.15

0 -.20
1/1/02
1/1/03
1/1/04

1/3/05

1/2/06
1/1/07
1/1/08

1/1/09
1/1/10

1/3/11

1/2/12
1/1/02
1/1/03
1/1/04

1/3/05

1/2/06
1/1/07
1/1/08
1/1/09
1/1/10

1/3/11

1/2/12

PX RPX

Graph 1. Daily Prices of PX and RPX

Source: Reuters
Table 1 summarizes the descriptive statistics of PX and RPX, which have negative skewness and high positive kurtosis.
These values signify that the distributions of the series have a long left tail and leptokurtic form. Jarque-Bera (JB)
statistics reject the null hypothesis of normal distribution at 1% level of significance.

Table 1. Descriptive Statistics

RPX
Mean 0.000210
Median 0.000721
Maximum 0.123641
Minimum -0.161855
Std. Dev. 0.015431
Skewness -0.524060
Kurtosis 15.43870
Jarque-Bera 18821.75
Probability 0.000000
Sum 0.609240
Sum Sq. Dev. 0.690062
Observations 2899

2 See: http://www.pse.cz/dokument.aspx?k=Exchange-Indices for description of PX

14
Applied Economics and Finance Vol. 2, No. 1; 2015

Table 2. Shows Augmented Dickey-Fuller (ADF) Test Results And Concludes That RPX Is Stationary.
Without Trend With Trend
Variable ADF stat p ADF stat p
RPX -39.7972*** 0.0000 -39.8306*** 0.0000
Note: *** denotes significant at 1% level.

We have also tested the mean model for an ARCH effect with the ARCH-LM Test. Table 3 shows ARCH-LM test
results. If the value of the test statistic is greater than the critical value from the distribution, the null hypothesis is
rejected indicating that there is no ARCH effect in the model.
Table 3. ARCH (1) LM Test Results
Dependent Variable of Model ARCH(1)LM Stat P
RPX 429,7907*** 0.0000

Note: *** denotes significant at 1% level.

We have employed ARCH, GARCH, GJR-GARCH and EGARCH processes using both Student-t and Generalized
Error distributions in addition to Normal distribution. Results show that strong GARCH and GJR-GARCH effects are
apparent for returns (Table 5, Table 6 and Table 7). The sum of the coefficients α and β are less than 1 for all different
distributions using GARCH and GJR-GARCH models. Nevertheless, the estimation of α is smaller than the estimation
of β in both cases, which reveals that negative shocks haven’t a larger effect on conditional volatility than the positive
shocks of the same magnitude. In GJR-GARCH model   0 , which means that the news impact is asymmetric or in
other words, bad news increase volatility. In the E-GARCH model negative and significant leverage effect parameter
indicates the existence of the leverage effect in returns. In EGARCH model, if GED parameter(r) value is 2 reveals a
normal distribution or a leptokurtic distribution if it is less than 2. In all models r is less than 2 and statistically
significant, which indicates that RPX is leptokurtic. This result is consistent with the skewness values reported in Table
1.
After all, ARCH effect is tested in all models. The null hypothesis that there is no ARCH effect cannot be rejected in all
models as shown in Table 5, 6 and 7. In the case of GJR-GARCH model with normal distribution, the null hypothesis is
rejected at 10% level of significance.
The in-sample evidence provides the history performance of the models. We have estimated variance for all models for
full sample period using static forecast. Then, we compared the forecasting performance of models which have been
used in this research. We have considered four statistics for evaluating the forecasting accuracy, which are employed in
the research of Wang and Wu (2012). These four measures are the Mean Square Error (MSE) and the Mean Absolute
Error (MAE) shown in equations (12) to (15):

 
n
MSE1  n 1   2t  ˆ 2t
2
(12)
t 1

 
n
MSE1  n 1   t  ˆ t (13)
2

t 1

n
MAE1  n 1   2t  ˆ 2t (14)
t 1

n
MAE2  n 1   t  ˆ t (15)
t 1

Where:
n is the number of forecasts

 t2 is the actual volatility

15
Applied Economics and Finance Vol. 2, No. 1; 2015

ˆ t2 is the volatility forecast at day t.

Table 4 shows the forecasting performance of GARCH, GJR-GARCH and EGARCH models. Generally, we have found
that EGARCH model has the greatest forecasting accuracy according to MSE2, MAE1 and MAE2, with only MSE1
showing better performance for GJR-GARCH.
Table 4. Comparison Forecasting Performance of GARCH Models
Normal Distribution Student t Distribution GED
GARCH GJR-GARCH EGARCH GARCH GJR-GARCH EGARCH GARCH GJR-GARCH EGARCH
MSE1 6.30E-07 6.05E-07 6.25E-07 6.31E-07 6.08E-07 6.29E-07 6.31E-07 6.06E-07 6.27E-07
MSE2 0.000480 0.000478 0.000460 0.000478 0.000477 0.000459 0.000478 0.000476 0.000459
MAE1 0.000243 0.000239 0.000234 0.000242 0.000239 0.000234 0.000242 0.000239 0.000234
MAE2 0.015476 0.015387 0.015342 0.015447 0.015398 0.015321 0.015439 0.015387 0.015315
Note: The values in bold face refer to smallest loss.
4. Conclusion
The article has examined three GARCH models namely GARCH, GJR-GARCH and EGARCH using three different
distributions, the Normal distribution, Student-t and Generalized Error distribution, in order to compare their forecasting
power for the volatility of the returns of the PX index for the Czech Republic stock market.
The results have shown that significant ARCH and GARCH effects are present in the data, which indicates that the
volatility in PX index returns are characterized by significant persistence and asymmetric effects. These results are
consistent with the results of Rockinger and Urga (2000), Haroutounian and Price (2010) and Vošvrda and Žikeš (2004)
which have used similar GARCH type models for the Czech stock market.
Finally, we have compared the in sample forecasting performance of the nine models for the period
08.01.2001-20.07.2012. The evidence shows that EGARCH model has the best forecasting performance based on real
statistical data.
References
Andersen, T. G., Bollerslev, T., & Lange, S. (1999). Forecasting Financial Market Volatility: Sample Frequency
vis-a-vis Forecast Horizon, Journal of Empirical Finance, 6(5), 457-477.
Bollerslev, T. (1986). Generalized Autoregressive Conditional Heteroskedasticity, Journal of Econometrics, 31(3),
307-327.
Bollerslev, T. (1987). A Conditional Heteroskedastic Time Series Model for Speculative Prices and Rates of Return,
Review of Economics and Statistics, 69(3), 542–547.
Brooks, C. (2008). Introductory Econometrics For Finance: Second Edition, Cambridge University Pres.
Emerson, R., Hall, S., & Zelweska-Mitura, A. (1997). Evolving Market Efficiency with an Application to Some
Bulgarian Shares, Economics of Planning, 30(2-3), 75-90.
Engle, R. F. (1982). Autoregressive Conditional Heteroskedasticity With Estimates of the Variance of UK İnflation,
Econometrica , 50(4), 987–1008.
Glosten, L. R., Jagannathan, R., & Runkle, D. (1993). On the Relation Between the Expected Values and the Volatility
of the Nominal Excess Return on Stocks, Journal of Finance, 48(5), 1779–1801.
Hájek, J. (2007). Czech Capital Market Weak-Form Efficiency, Selected Issues, Prague Economic Papers, 4, 303-318.
Haroutounian, M. K., & Price, S. (2001). Volatility İn The Transition Markets of Central Europe, Applied Financial
Economics, 11, 93-105.
Mittnik, S., Paolella, M. S., & Rachev, S. T. (2002). Stationarity of Stable Power-GARCH Processes, Journal of
Econometrics, 106, 97–107.
Nelson, D. B. (1991). Conditional Heteroscedasticity in Asset Returns: A New Approach, Econometrica, 59(2),
347-370.
Rockinger, M., Urga, G. (2000). Time Varying Parameters Model to Test for Predictability and Integration in Stock
Markets of Transition Economies, CERP Discussion Papers 2346.

16
Applied Economics and Finance Vol. 2, No. 1; 2015

Scheicher, M. (1999). Modeling Polish Stock Returns. in Helmenstein, C. (ed.), Capital Markets in Transition
Economies. Cheltenham, Edward Edgar, UK, 417–437.
Scheicher, M. (2001). The Comovements of Stock Markets in Hungary, Poland and the Czech Republic. Internatıonal
Journal of Finance and Economics, 6, 27–39.
Shields, K. (1997). Stock Return Volatility on Emerging Eastern European Markets. The Manchester School Supplement,
118–138.
Syriopoulos, T. (2007). Dynamic Linkages Between Emerging European And Developed Stock Markets: Has The Emu
Any Impact, International Review of Financial Analysis 16, 41-60.
Taylor, S. J. (1986). Modelling Financial Time Series, John Wiley & Sons, Chichester.
Thalassinos, I. E., & Politis, D. E., (2011). International Stock Markets: A Cointegration Analysis, European Research
Studies Journal, XII (2), 113-130.
Thalassinos, I. E., Maditinos, D., & Paschalidis, A. (2012). Observing Evidence of Insider Trading in the Athens Stock
Exchange (ASE), Journal of Economic Structures, 1(8), Dec.
Thomas, S. & Mitchell, H. (2005). GARCH Modeling of High-Frequency Volatility in Australia’s National Electricity
Market, Discussion Paper, Melbourne Centre for Financial Studies.
Vošvrda, M., & Žıkeš, F. (2004). An Application of the GARCH-t Model on Central European Stock Returns, Prague
Economic Papers, 1, 26-39.
Wang, Y., & Wu, C. (2012). Forecasting Energy Market Volatility Using GARCH Models: Can Multivariate Models
Beat Univariate Models, Energy Economics.
Yu, J. (2002). Forecasting Volatility in the New Zeal& Stock Market, Applied Financial Economics, 12, 193-202.

Appendix

Table 5. Normal Distribution

GARCH(1,1) GJR-GARCH E GARCH
Value p Value p Value p
Mean equation
0 0.0009 0.0000 0.0006 0.0023 0.0005 0.0055
Variation equation
0 4.69E-06 0.0000 6.07E-06 0.0000 -0.5147 0.0000
 0.1318 0.0000 0.0727 0.0000 0.2531 0.0000
 - - 0.1036 0.0000 -0.0687 0.0000
 0.8496 0.0000 0.8441 0.0000 0.9636 0.0000
AIC -5.9192 -5.9291 -5.9282
SIC -5.9110 -5.9188 -5.9179
DW-stat 1.8863 1.8888 1.8890
ARCH-LM (1) Test p-value of Chi Sq 0.306626 0.0846 0.1493
Obs. 2899

17
 Applied Economics and Finance Vol. 2, No. 1; 2015

Table 6. Student’s T Distribution

GARCH(1,1) GJR-GARCH E GARCH
Value p Value p Value p
Mean equation
0 0.0009 0.0000 0.0008 0.0000 0.0008 0.0001
Variation equation
0 4.34E-060.00005.36E-060.0000-0.47000.0000
 0.1192 0.0000 0.0696 0.0000 0.2354 0.0000
 - - 0.0895 0.0000 -0.613 0.0000
 0.8617 0.0000 0.8563 0.0000 0.9673 0.0000
AIC -5.9526 -5.9583 -5.9577
SIC -5.9423 -5.9459 -5.9453
DW-stat 1.8855 1.8872 1.8873
ARCH-LM(1) Test
0.3879 0.1387 0.2596
p-value of Chi Sq.
Obs. 2899
Table 7. Generalized Error Distribution(GED)

GARCH(1,1) GJR-GARCH E GARCH

Value p Value p Value p
Mean equation
0 0.0009 0.0000 0.0007 0.0001 0.0007 0.0002
Variation equation
0 4.56E-06 0.0000 5.72E-06 0.0000 -0.4947 0.0000
 0.1256 0.0000 0.0712 0.0000 0.2446 0.0000
 - - 0.0960 0.0000 -0.0646 0.0000
 0.8546 0.0000 0.8495 0.0000 0.9653 0.0000
r 1.5083 0.0000 1.534154 0.0504 1.5329 0.0498
AIC -5.9427 -5.9493 -5.9484
SIC -5.9324 -5.9369 -5.9360
DW-stat 1.8856 1.8876 1.8878
ARCH-LM (1) Test p-value of Chi Sq. 0.2896 0.074 0.1979
Obs. 2899

This work is licensed under a Creative Commons Attribution 3.0 License.

18

View publication stats